Should digital headroom for DACs and DSPs be tested as part of review measurements?

[etc6849]

This is a great video covering DAC and DSP distortion for interpolator overload distortion and intersample overs (surprisingly present in 1000's of CDs and can show up 1000's of times in a track).

Also, if you think lowering the digital volume always fixes this issue, it may not! It comes down to if the digital volume reduction is done before or after interpolation. About 8-9 minutes in is a MiniDSP software update discussion where Benchmark identified an issue and MiniDSP, who was very receptive, did a firmware fix to prevent.

 

[JSmith]

Search results for query: overs

www.audiosciencereview.com

JSmith

 

[staticV3]

dBTP levels above 0dBFS are not compatible with proper mastering practices IMO, unlike what John Siau appears to claim.

Yes, problematic tracks exist. No, they're not the majority.

But so what if a track has 1000 samples above 0dBFS? A typical track consists of around 10M samples. So 0.01% of samples would be clipped.
That's a total of 20 milliseconds of clipped audio scattered across an entire track, or this long (in sum):

Most of all though: Where are the ABX tests proving audibility of ISOs in musical material?

Yes, if I had the choice then of course I'd like my DAC to also handle this edge case gracefully.

But is it ThE BiGGeSt PrObLeM in DiGiTaL AuDiO??! Not even close, at least not until we've solved the loudness war.

 

[AgarWave]

I think this would be useful if it is treated as a small, repeatable stress test rather than a single pass/fail number. A review could report the maximum clean true-peak level at several frequencies, for example +0.5, +1, +2 and +3 dBTP, then repeat the test with the device's digital volume reduced by 3 dB.

That second run matters because it can distinguish attenuation before interpolation from attenuation applied later in the signal path. For DSP products, the same test could be repeated with a unity-gain filter and with a positive-gain EQ preset, because internal headroom may differ from the DAC's reconstruction headroom.

The result would not imply that intersample overs are the dominant audible problem in normal music, but it would show whether a DAC or DSP fails gracefully, clips abruptly, or changes behaviour depending on where volume control is implemented.

 

[Chris A]

Wow...

This also tracks with my experiences demastering music tracks, part of which is reducing the music track output levels to a ReplayGain level of zero (whereas the as-released tracks are typically -6 or even -9 dB--compressed and clipped tracks). Everything sounds better when the "zero" ReplayGain level is used across all the DSPs and D/As, giving typically more than 3 dB of margin from zero dB (maximum gain).

And I agree with Mr. Siau, I do like the sound of DSD over PCM tracks--but probably the reason is that DSD tracks cannot be "mastered" in the typical sense without first being converted to some form of PCM internally (like the old Sonoma system for DSD mastering was doing without users actually understanding what was occurring). What this inability to master DSD streams actually does is give the users a much better chance of achieving phase fidelity with the original recording, since the mastering guys can't fool with the digital signal. I've found independently that phase distortion really is audible (as Toole also has pointed out in the various versions of his book).

Thanks for the heads up. One would have thought that this kind of problem would have been caught years ago by DAC and DSP manufacturers, especially in light of the Loudness War practices by the record companies pushing everything to the limit of loudness.

Apparently, it's been the "elephant in the room" but no one was paying attention.

Chris

 

[amirm]

There are two issues here:

1. How big of a deal this is.

2. What is the fix?

In all of my experience playing heaven knows how many pieces of music, I have never, ever heard such an artifacts. I have once, verified that my son had a repeatable case that triggered this on one DAC, and not others. At least I think that was the case.

Now to #2. The only fix available is to assume such clipping exists AT ALL TIMES, and therefore, steal dynamic range from everything played. That is, reduce the max level of the incoming digital signal, and then increase gain in analog domain to compensate, which adds constant noise.

But let's say you want to do that. As John says, media players and streams already have volume control performed in digital domain where you can dial in attenuation. And if you are going to do EQ, then you will have to do that anyway. Indeed, this is why Roon player has dynamic range compensation as a checkbox:

See how I have enabled "headroom management?" That is a constant attenuation which you can dial to any value, include the recommended 3 dB for intersample overs.

Further, if you have just about any DAC that doesn't say otherwise, with a volume control, that volume control is in digital domain and prior to actual conversion, sample rate conversion, etc. To with, here is the ESS DAC:

This is what I have on my desk and since I never play at 0 dBFS, I am covered. And what I recommended people to buy, a DAC with volume control so you don't need a pre-amp either.

If one wanted a "proper" fix, it would require a tool that scans every track and dials in a headroom. Media players could do this easily if this were a real problem. Hardware DACs cannot do this since they don't have memory to store entire tracks, analyze them in advance, and then play.

The only time you may want such a thing would be if you still operated a CD player, fed it to a DAC with no volume control, and then a pre-amp. Since hardly any of us do this, I am not about to force this down the throat of every DAC to reduce its digital output.

So let's not keep bring this up. Benchmark and a few others who do this should have a switch that defeats it and with it, gain another 3 dB of SINAD.

 

[KSTR]

A full-featured DAC characterization should check ISO behavior because it can be anything from benign and inaudible to nasty and chaotic, like wraparound etc. The exact reaction to ISOs in the source material depends on many variable factors, like choice of DAC filter, notably any ASRC processing, the analog circuit, etc. That makes complete testing an elaborate task, even if there were an agreement of what would be good, standardized test signals.

Anyway, I think it would feasible to at least show the reaction to the well-known fs/4 +3dBTP test signal (as shown in the video), select the steepest/longest digital filter and all volume controls set to 100%. Spectrum and waveform would give insights.

 

[KSTR]

Most of all though: Where are the ABX tests proving audibility of ISOs in musical material?

We don't have ABX showing the relevance of 130dB SINAD vs. 100dB either, still DACs are tested for, and ranked by, SINAD here. Thus, not a good argument.

 

[Purité Audio]

FUD it’s a common theme in Benchmark’s marketing.
Keith

 

[antcollinet]

We don't have ABX showing the relevance of 130dB SINAD vs. 100dB either, still DACs are tested for, and ranked by, SINAD here. Thus, not a good argument.

We do have data though about the thresholds of audibility for noise and distortion. And those go up above 100dB for some peoople. The SINAD ranking enables us to rule out a whole raft of devices that might have audible noise and distortion according to our own judgment of what might be audible for us as individuals. Personally for me (like presumably you) I am very comfortable with 100dB - probably also 90.

For ISOs I am not aware of any study that demonstrates audibility at any level in real music. All I have is my own personal experience of never having heard anything I could attribute to an ISO. In the absence of better scientific data telling me otherwise - the only conclusion I can reach is: It doesn't matter so why waste time and energy measuring it?

Perhaps a test for "faulty" behaviour such as wrap arounds might be worth doing which could then be another "exclude this device" metric.


EDIT:

This is what I have on my desk and since I never play at 0 dBFS, I am covered. And what I recommended people to buy, a DAC with volume control so you don't need a pre-amp either.

Exactly

 

[KSTR]

Perhaps a test for "faulty" behaviour such as wrap arounds might be worth doing.

That's exactly what I'm proposing. A quick and feasible check to weed out the crap.

 

[KSTR]

For ISOs I am not aware of any study that demonstrates audibility at any level in real music.

I once published files here with emulated ISO behavior (of the simple kind, clipped vs. not clipped) and some seem to have heard it.

 

[amirm]

A full-featured DAC characterization...

That's not the business I am in. I am not running a lab here.

Every additional measurement obscures the overall message of the review for all but a few. The more measurements, the worse this gets. There is a reason we are successful in such testing and others who publish books by running scripts and such, are not.

 

[MaxwellsEq]

Content producers should avoid Inter-Sample Overs. It should not be necessary for DAC manufacturers to add headroom to compensate for a content mastering failure.

But ISOs are possibly inaudible. They may be the least audible of the countless mistakes very audible made during mastering and remastering.

Domestically, they are easily avoidable if the DAC has onboard digital attenuation (and / or DSP); simply never run the DAC at 0dBFS.

If the DAC has a fixed output, it could be designed with 2 to 3 dB of headroom. This would negatively affect its measured SINAD making it look worse than other products by 2 to 3dB. But if the device is in the -110dB or better performance range, the disadvantage would be inaudible, and would mitigate the possibility of ISOs being audible. For me this makes good sense.

 

[Chris A]

Perhaps just adding a little time for independent voices to characterize the extent of this newly discovered "panic" elsewhere is the likely course of events, no matter what initial reactions might be.

I know that I'm not going to run for the exits just yet, seeking immediate fixes. I'm going to keep on the same pathway as before this "revelation", reading intently but also using my ears and following my nose. These two have served me well in the past.

Chris

 

[MaxwellsEq]

Perhaps just adding a little time for independent voices to characterize the extent of this newly discovered "panic" elsewhere is the likely course of events, no matter what initial reactions might be

It's not especially new. I remember discussing it in the 1980s regarding professional ADCs, although it was considered a mathematical rather than real problem. I don't recall it being called "Inter-Sample Overs". It was considered good sense to leave a few dB in hand to avoid the problem.

At the time I felt there was a "habit and practice" issue. Analogue tape could be pushed into the red and some liked the audible compression effect. Doing so also reduced audible tape hiss. With ADCs pushing into the red sounded terrible, and music producers now missed out on a free compression artefact. But with ADCs giving significantly lower noise floors, there was no need to push into the red

 

[Chris A]

Analogue tape could be pushed into the red and some liked the audible compression effect. Doing so also reduced audible tape hiss. With ADCs pushing into the red sounded terrible, and music producers now missed out on a free compression artefact. But with ADCs giving significantly lower noise floors, there was no need to push into the red

Someone I think needs to tell the mastering guys that digital isn't analog, and that hard limits to gain structure need to be respected with margin control (otherwise known as "design factors" for non-digital domains).

I know that the reasons for the mastering types pushing all of their work products to the limit of loudness has its roots in...well...non-engineering reasoning.

Chris

 

[Mike123]

If this problem exists in any musical material, it's not its worst problem. With a probability approaching 1, this material is compressed to the limit and is difficult to listen to precisely for this reason.

 

[MaxwellsEq]

I've not scanned all my music, but a good percentage, perhaps 20% has Inter-Sample Overs. Quite a lot of that is not highly compressed.

To be clear, Inter-Sample Overs are NOT (necessarily) a consequence of the loudness wars. You can have a recording crushed to DR3, but as long as the peaks are 3 to 4dB below 0dBFS, no ISOs will be embedded.

 

[KSTR]

That's not the business I am in. I am not running a lab here.

I know, and that's what I was trying to get across. A simple check with the fs/4 full-scale signal could be useful but it is totally up to you if you would add it and omit some other tests in exchange, like the linearity test which is useless with any delta-sigma DAC (which by design don't have linearity issues).